基于高斯混合模型的地铁牵引系统健康度评估

Health Evaluation of Metro Traction System based on Gaussian Mixture Models

文章基于高斯混合模型,采用数据挖掘技术,对地铁牵引系统健康度评估进行研究。主要研究方法为利用无监督学习和有监督学习两种方式,分别建立机器学习模型。首先提取原始特征,采用方差过滤法和主成分分析法进行特征降维;其次在无监督学习下对是否工况分离进行讨论,选用高斯混合模型将得到的类概率值作为设备健康度评分;最后在有监督学习下选用逻辑回归模型评估牵引系统的健康度。结果表明,在无监督学习工况分离的情况下,高斯混合模型的表现状态有较高的预测精度和模型准确度。

By using data mining techniques,a health evaluation of the traction system for metro vehicles is studied on the basis of Gaussian mixture models.With both unsupervised learning and supervised learning as the main research methods,machine learning models are established respectively.Firstly,we extract primitive characters with the aid of variance filtration and main components analysis for feature dimension reduction;secondly,we discuss on the separation of working conditions in unsupervised learning mode,and take categorized probability value achieved from Gaussian mixture models for grading of equipment health;and lastly,we choose logic regression models in supervised learning mode to evaluate the health of the traction system.Research results show that Gaussian mixture models with separated working conditions in unsupervised learning mode present higher prediction accuracy and model accuracy.